Changelog

New Features

• •

  🖼️ Better Image Support: Send images straight into a conversation and let AI read a screenshot, mockup, or photo before it acts. Image generation moves to a stronger engine with reference-image editing and steadier results; images can be larger and sharper (32MB each), and a failed generation now tells you why. Drop in a screenshot and say "three more in this style."

• •

  🧭 Ask AI How to Use It: Not sure where a feature or setting lives? Just ask. AI takes you straight there and highlights it, tells you the current state — your language, credits left — and can file feedback when something's off. Ask "where do I turn on auto-allow permission" and it'll take you right there.

Improvements

• • 📡 More Reliable Messaging Channels: While AI is busy, new channel messages queue up and get answered in order once it's free — and survive a restart. AI also picks up the channel and conversation context, so replies fit the situation better.
• • After cancelling a generation midway, the conversation stays stable and won't break from the interruption

This release marks a positioning upgrade: MCPlato evolves from an "AI Workspace" into an "AI Partner," and we've rewritten the desktop experience and runtime engine around it.

AI Partner

• • Every workspace is a teammate: each has its own name, avatar, memory, and area of expertise. Your relationship with AI shifts from "using a tool" to "working with a colleague." For example, keep a "weekly report partner" around to compile meeting notes for you every week.
• • A home for your whole team: a new persistent home window shows each partner's status, unread messages, and update progress at a glance, and stays one click away in the tray.
• • Manage partners right from the list: right-click any partner to rename, change avatar, or delete — no need to open settings.

Infinite Desktop

• • Multi-window mode: switch between "single window" and "multi-window" in settings. In multi-window mode, every partner gets its own window — spread them across as many screens as you have.
• • Drag a tab out to detach: drag any tab out of a window to spawn a new canvas; drag it back to merge automatically. Send your "weekly report" partner to a second screen while you keep talking design with another partner on the main one.
• • Merged tab indicators roll up: unread badges and pending confirmations on a merged tab now aggregate across all panes of the same partner — nothing slips past you.

Higher parallel bandwidth, faster AI response

• • Faster local inference engine, on by default for new sessions: startup, first token, and tool calls are all noticeably quicker.
• • True parallel work across partners: windows never block each other. While one partner runs a long task, the others keep going. Run two 30-minute jobs on two partners and chat with a third — nobody waits on anyone.
• • Long tasks no longer get cut off: as the context limit approaches, memory is reorganized automatically, key information is preserved, and the task runs through to completion.
• • Smarter permission auto-allow: tool permission auto-allow is now managed centrally by the main process — it stays in control across window switches, and dangerous operations are blocked automatically.
• • Tighter startup: auth state, configuration, and window positions load together — less flicker, and you land right back where you left off after sign-in.

New Features

• •

  Local Tools Runtime Management: A new "Local Tools" page in Settings lets you check install status, versions, and update progress for runtimes like Python, and FFmpeg, then install or update them in one place.

• •

  Built-in Python Runtime: Skills that rely on Python and pip dependencies can now run without manual system setup. If a skill needs extra packages, chances are it can just start working.

• •

  More Reliable Long-running Tasks: AI now handles longer, multi-step local workflows more steadily, so long conversations and complex tool chains are less likely to stall halfway through.

Improvements

• • Local command execution now follows the current working directory and environment variables more accurately, making multi-step flows more dependable
• • Permission prompts and sandbox behavior are clearer, with fewer unnecessary interruptions during repeated actions

New Features

• • Deep Lark integration: Added 29 new Lark tools, covering CRUD operations for Bitable, task and subtask management, group and direct messaging, document comments, cloud document operations, and global search. Just complete Lark OAuth authorization in ClawMode to use them. Try telling the AI: "Help me update last week's report data into Bitable," or "Send a message to the design group to follow up on the progress."
• • Support for sending any file in chat: The input box has been upgraded from "Add Image" to "Add File." PDFs, Excel files, code files—just drag them into the chat and send them to the AI directly.
• • Enhanced WeChat channel: Now supports receiving voice messages (automatically transcribed into text) and files. Images generated by the AI can also be sent back directly through WeChat.

Improvements

• • Scheduled tasks are now bound to the workspace that owns the conversation.
• • Channels now forward only tool execution results that include images, reducing redundant message pushes.

✨ New features

• • WeChat integration: Scan a QR code in ClawMode to log in to a WeChat Account. Messages sent by users through the bot will automatically enter the AI conversation flow, with support for receiving image messages 🖼️. No business verification is required; persona accounts are also supported.
• • Support for 10 languages: Added six new languages—Traditional Chinese, Spanish, German, French, Korean, and Portuguese 🌍. Together with the existing Chinese, English, Japanese, and previously supported languages, the interface is now available in 10 languages. You can switch languages in Settings, and all windows will update simultaneously.
• • Tool usage memory: When typing @ in the chat box to call a tool, recently used tools will appear at the top ⬆️, so you do not have to search for them every time.

🛠️ Improvements

• • The Virtual Assistant window and Workspace Settings dialog now support multilingual display.

✨ New Features

• • Phone Number Login — Supports quick login via SMS verification code, with no need to remember your email password. Switch to the "Phone Number" tab on the login page to use it.
• • Multilingual Support — The app interface now fully supports switching between Chinese and English. You can change the language in Settings - Account, and the AI conversation system prompt will be adjusted accordingly after switching.
• • WeChat Customer Support Channel — Connect a WeChat customer service account in ClawMode, and messages sent by users through WeChat will be automatically handled and replied to by AI.

🔧 Improvements

• • Improved global network connection stability
• • Message channels can now detect and notify users of unsupported media types instead of silently ignoring them

New Features

• • ClawMode Autonomous Work Mode: Workspaces now support a brand-new ClawMode standalone window. Claw can continuously monitor the current work state in the background and step in proactively when needed, evolving AI from simply “waiting for instructions” to actively “participating in the workflow.”
• • ClawMode Diary System: ClawMode workspaces now automatically create clawmode/diary.md for logging and reviewing work progress. Each time a session starts, the system also automatically loads the previous day’s diary as context, helping Claw maintain a more continuous understanding of your tasks and working rhythm.
• • QQ Bot Channel Integration: Added support for the official QQ bot adapter, with image send/receive support, Markdown message formatting, and complete getting-started guides in six languages to help developers integrate faster.
• • Browser Automation (Agent Browser): Built-in agent-browser browser automation now replaces Playwright MCP. Chromium is installed automatically on startup with no manual setup required.
• • macOS Intel Version Support: The official website now also provides a macOS Intel (x64) download option, allowing Intel Mac users to download and use a native version directly.
• • Ask-User Question Forwarding Across Channels: Questions initiated by AI during conversations through the AskUserQuestion tool can now be forwarded to channels such as Feishu. Users can reply directly within the channel, and their answers will be automatically injected back into the conversation flow without needing to switch back to the app.

Improvements

• • Scheduled Task Enhancements: Workflows can now automatically stop and deliver results once AI completes a task. When a workspace or session no longer exists, orphaned tasks are automatically paused instead of being silently skipped. In addition, the settings panel now receives task events in real time and refreshes automatically without manual reloading.
• • Channel Permission Label Optimization: The original labels “Allow All” and “Deny All” have been updated to the clearer “Full Access” and “Safe Mode,” making permission settings easier to understand at a glance.
• • Feishu Integration Guide Update: The Feishu setup guide has been fully rewritten, now including instructions for bulk importing permissions via JSON, along with a full set of updated high-resolution screenshots aligned with the latest official documentation.
• • Embedded Node.js Upgraded to v22.21.1: The built-in Node.js runtime has been upgraded to v22.21.1, keeping it aligned with the current development environment. The runtime version used for skill scripts has also been updated accordingly.

Bug Fixes

• • Fixed an occasional issue with automatic updates on Windows.
• • Fixed a silent error that could occur when an AI session ID became invalid.
• • Fixed an issue where channel messages could be delivered repeatedly.

🦞 Claw Mode

• • Introduced Claw Mode, enabling MCPlato to run continuously 24/7.
• • Supports integration with multiple IM platforms, including:
  • • Telegram (TG)
  • • Feishu (Lark)
  • • Slack
  • • Discord
• • Enables cross-platform message processing and automated workflows, allowing MCPlato to continuously handle tasks and messages.

🎨 AI Image Generation & Editing (Nano Banana 2)

• • Integrated Nano Banana 2 for image generation and editing.
• • Supports chat-based image generation — simply describe what you need to create images.
• • Supports AI-powered image editing to modify, enhance, or remix existing images.
• • Ideal for design drafts, marketing materials, and content creation.

---

🛠 Improvements

• • Improved overall system stability and runtime performance.
• • Optimized task scheduling and message processing logic.
• • Enhanced parts of the UI interaction for a smoother experience.

---

🐞 Bug Fixes

• • Fixed issues causing instability during certain task executions.
• • Fixed compatibility issues with some IM integrations.
• • Resolved several known stability-related bugs.

New Features

• •

  Dynamic Theme System: Supports three theme modes—Light, Dark, and High Contrast. Theme settings are synced across all windows. Switch in Settings > Appearance, or use the system default to follow your OS automatically.

• •

  Scheduled Task Management Panel: Adds a visual interface for managing scheduled tasks, supporting creation, editing, and deletion of recurring tasks. Use cron expressions to define run times, with task status clearly visible at a glance.

• •

  Sprite Assistant Enhancements: The input area now supports auto-expanding multi-line text and drag-and-drop image input. Hover over the bottom 40% of the screen to quickly summon the input box.

• •

  WebM Transparent Video Support: The Sprite Assistant avatar now supports WebM format, delivering smoother and more natural animations.

• •

  Embedded FFmpeg Runtime: Built-in FFmpeg tools enable audio/video processing directly in chat, with no additional installation required.

• •

  Image Generation & Editing: Generate, edit, and composite images directly in conversations. Describe what you want—AI will create an image or modify an existing one.

• •

  Kimi Model Support: Adds support for the Moonshot Kimi model series, fully compatible with reasoning content and tool-calling capabilities.

• •

  Gemini Model Support: Adds support for Google’s Gemini model family, including the latest models such as Gemini 3 Flash—now available alongside Claude and GPT as the three major model families.

• •

  Web Search Tool: Built-in web search capability allows AI to retrieve real-time online information to assist with answers.

• •

  Automatic Permission Countdown: Adds an auto-confirm permission option in Settings. You can set permission requests to be automatically allowed after 10/20/30 seconds; hovering over the popup pauses the countdown.

• •

  Mid-Cycle Subscription Upgrades: Upgrade your plan anytime during the subscription period. The price difference is prorated based on remaining days and takes effect immediately.

Improvements

• •

  File Browser Performance Optimization: Only monitors expanded directories, greatly reducing system resource usage. Expanding the file tree on desktop or in large directories is no longer laggy.

• •

  Automatic Windows Environment Setup: Automatically configures the embedded Git path, so terminal commands work out of the box.

• •

  AI Avatar Display Optimization: Adds an AI avatar to the chat UI and shows it when a new turn begins, making it easier to distinguish message sources.

• •

  Real-Time Subscription Status Display: The title bar now shows the current subscription status so you can always see your plan at a glance.

• •

  Settings Page Optimization: Adjusts the loading order of the custom model settings page after login, eliminating brief flicker.

Bug Fixes

• • Fixed an issue where the chat scroll position would jump when using Chinese IME input.
• • Fixed an issue where draft content in the input box could be lost after switching workspaces.
• • Fixed an issue where the AI avatar could get stuck during some streaming conversations.
• • Fixed an issue where the subscription status could occasionally fail to load on app startup.
• • Fixed an issue where the input area height could display incorrectly after restoring drafts.
• • Fixed a potential crash during macOS auto-update restart.

New Features

• • Sprite Session Assistant: Introduced a desktop floating AI assistant, allowing users to start conversations via the bottom input area or by clicking the avatar.
• • Unified @ Mention System: The conversation input now supports @ to quickly invoke skills and MCP tools. A new Tools tab consolidates skills and MCP tools with search support and color-coded icons. Try typing @image to invoke image generation.
• • Image Generation and Editing: Generate, edit, and compose images directly within conversations. Describe the scene you want, and the AI will create or modify images, supporting multiple styles and resolutions.
• • Gemini Model Support: Added support for Google Gemini models, including the latest Gemini 3 series, now available alongside Claude and GPT as one of the three major model families.
• • Kimi Model Support: Added support for Moonshot Kimi k-2.5 series models, with full compatibility for reasoning content and tool invocation.
• • Web Search Tool: Built-in web search capability enables the AI to retrieve real-time information to assist with responses.
• • Automatic Permission Countdown: Added an option in settings to automatically approve permission requests after 10/20/30 seconds. The countdown pauses when hovering over the permission dialog.
• • Mid-cycle Subscription Upgrade: Users can now upgrade their subscription plans at any time during the billing cycle. The price difference is prorated based on remaining days and takes effect immediately.

Improvements

• • File Browser Performance Optimization: Only expanded directories are monitored, significantly reducing system resource usage. Expanding file trees on the desktop or in large directories is now smooth and responsive.
• • Automatic Windows Environment Configuration: Automatically configures the embedded Git path, enabling terminal commands to work out of the box.
• • AI Avatar Display Optimization: Added AI avatars to the conversation interface, displayed at the start of each new turn for clearer message attribution.
• • Real-time Subscription Status Display: The current subscription status is now shown in the title bar for quick reference.

Bug Fixes

• • Fixed an issue where the conversation view would jump during Chinese IME input.
• • Fixed an issue where input drafts could be lost when switching workspaces.
• • Fixed an issue where AI avatars could become stuck during some streaming responses.
• • Fixed an issue where subscription status could occasionally fail to load on application startup.
• • Fixed an issue where the input area height could display incorrectly after restoring drafts.
• • Fixed a potential crash during automatic restart after updates on macOS.

New Features

• • Added support for @ mentions of Skills and MCP tools in conversations. Type @ in the input box to quickly search for and invoke installed skills or tools. Three category tabs are available: Files, Tools, and Conversations.
• • Try typing @image to invoke the image generation skill.

Improvements

• • The file browser now only listens to expanded directories, significantly reducing system resource usage. Expanding file trees on the desktop or in large directories no longer causes lag.
• • Improved draft behavior in the input box when switching workspaces.
• • Optimized the performance of the AI avatar during streaming conversations.

Bug Fixes

• • Fixed an issue where the conversation view would jump during Chinese IME input.
• • Fixed an issue where the input area height displayed incorrectly after restoring drafts.

New Features

• • Windows users can now use terminal commands directly—the app includes a complete Bash environment.

Improvements

• • Added AI avatar to the chat interface, displayed at the start of new conversation turns for easier message source identification.
• • Optimized streaming message creation timing to avoid showing empty messages when only tool calls are present.
• • Improved log collection logic in the feedback feature for more complete issue context capture.
• • Skill tools are now stored in directories by skill name, allowing multiple skills to use tools with the same name without conflicts.

Bug Fixes

• • Fixed an occasional authentication error on startup.
• • Fixed empty messages appearing when AI responses contained only tool calls.
• • Fixed an issue where feedback couldn't properly collect logs in certain cases.
• • Fixed model name not being correctly replaced in streaming responses.

New Features

• • Subscription status now displays in the title bar. Open the "Account" panel in Settings to view detailed subscription information and remaining credits.
• • New users receive starter credits upon registration, so you can start exploring right away without subscribing first.
• • Added contact information to the "About" panel in Settings—if you run into any issues, you can email us directly.

Improvements

• • Simplified the input area by removing the voice input button and detailed token usage display for a cleaner interface.
• • Error messages are now more user-friendly and no longer show technical log file paths.
• • Fixed an occasional login state issue that could occur at app startup.

New Features

Windows System Tray When minimized, the app will be tucked into the system tray. Double-click the icon to quickly bring up the main window.

Login Optimization Login now uses browser-based authorization that automatically returns to the app, no manual action required.

Update Notifications When important updates are available, the app will display an upgrade prompt and guide you through the update.

Improvements

• • Account settings now display usage information, keeping track of remaining quota at any time
• • Right-click on a tab to quickly rename it
• • File tree supports directory collapsing
• • Window position memory is more reliable

Fixes

• • Fixed data conflicts caused by running multiple app instances